Generative AI Tools:

Generative AI platforms allow users to create content, ranging from text to images, using artificial intelligence models. While there are several AI tools available, Instructional Designers should focus on using Microsoft Copilot when possible in guides, consultations, workshops, and other professional development activities. At the same time, designers must recognize that some faculty may prefer to use a different tool.

Microsoft Copilot

• Previously known as Bing Chat.
• Generative language model with text based prompts.
  • Get detailed, human-like responses with footnotes linking back to the original sources.
  • Provides up-to-date information (since it is connected to the internet), which is a capability ChatGPT’s free version does NOT boast.
  • Get help with creative tasks, such as writing a poem, essay, or song.
  • Solve complex math or coding tasks.
  • Generate images from text.
• Uses ChatGPT 4 with some modifications.
  • Copilot has three settings: Creative, Balanced, and Precise.
    • Creative setting, as the name implies, provides more creative responses at the expense of accuracy.
    • Precise provides more accurate but less creative responses.
    • Balanced setting is a ‘middle-ground’ setting which provides more creativity than the Precise setting, but more accuracy than the Creative setting.
• Uses DALL-E 3 for image creation.

Microsoft offers three versions of Copilot:

UVU currently pays for Copilot. Copilot Pro is intended for individuals and not really for UVU use. Copilot for Microsoft 365 can be purchased by UVU departments. It has not been purchased at this time.

• Ways to access MS CoPilot:
  1. Web page.
  2. Web extension for the Microsoft Edge web browser.
     • Unfortunately, it is only available for the Microsoft Edge web browser. The output of CoPilot should be similar regardless of whether you use the web page or browser extension.
  3. Mobile apps are unavailable right now. They will be available if OTL purchases Copilot for Office 365.
• Provides privacy when logged in using a UVU account.
  • Prompts and content uploaded into Copilot are not used to train its model. Prompts and content are deleted after 30 days.

OpenAI ChatGPT

• Most popular generative AI platform.
  • Because Microsoft Copilot uses ChatGPT, the capabilities and functionality are similar.
• Free tier uses version 3.5, which is often inferior to other tools.
  • The paid tier uses ChatGPT 4.0 Turbo which is considered to be vastly superior over version 3.5. Some departments may choose to pay for an enterprise license to ChatGPT.

Google Gemini

• Gemini is Google’s conversational AI chat service. It is meant to function similarly to ChatGPT, with the biggest difference being that Google’s service will pull its information from the web. Like most AI chatbots, Gemini can code, answer math problems, and help with your writing needs. It also has the ability to generate images.
• Gemini is free but also has “Advanced” tier for additional cost. UVU does not have an agreement with Google to use Gemini Advanced. 

Anthropology Claude

• is preferred by some faculty.
• markets itself as a more socially conscious generative AI tool.
• allows for uploading longer documents and may be a better fit than Microsoft Copilot.
• is free to use with a limited number of prompts permitted per hour. There is a paid tier ($20 per month) as well.

Jukebox by OpenAI:

• Jukebox is a generative model that creates music in various styles and genres. It’s designed to generate novel music compositions.

Resources

Pluralsight courses on generative AI

Prompt Engineering

Prompts for AI are a critical part of getting the results you want. It is important to provide the full context of your request when prompting AI. Here are some ideas that may help you create successful prompts.

PREP

• Prompt: Introduce the question with a prompt
• Role: Give it a role or voice
• Explicit: Be explicit in your instructions
• Parameters: Set the parameters of the answer

EDIT

• Evaluate your AI output content for language, facts, and structure
• Determine accuracy and corroborate with sources
• Identify biases and misinformation in output.
• Transform content to reflect adjustments and new findings

From the book: “The AI Classroom: The Ultimate Guide to Artificial Intelligence in Education” by Dan Fitzpatrick, Amanda Fox, Brad Weinstein, copyright 2023

Using Generative Visual AI Tools within Course Development

Before using a visual AI tool, first ask yourself the following questions:

1. Does your content need to be supported by visual aid? If not, there’s no reason to use a visual AI tool.
2. Does Adobe Stock Photos have something to fit the bill? If it does, there’s no reason to use a visual AI tool.
3. How complex is your scenario? If your scenario is relatively simple, you’re more likely to have luck using a visual AI tool. Contrastingly, if your scenario is relatively complex, you’re more likely to have difficulties getting a visual AI tool to create the image you want.

 If you have determined that an AI tool is right for your situation, you need to decide which AI tool to use. UVU provides two visual AI tool options for employees: Microsoft Copilot and Adobe Firefly. With UVU’s license, Copilot offers unlimited searches while Adobe Firefly has an individual account limit of 525 credits per month. These credits apply not only to “searches” in Adobe Firefly, but also to generative fill in Photoshop and Illustrator.

Copilot

We recommend that you start with Copilot. To use it, first sign into the web version of Copilot with your UVU credentials. Select your conversation style (creative, balanced, or precise). Then, type a prompt in the search box at the bottom of the screen. As you craft your prompt, keep in mind that clear and simple prompts with specific details may generate better results.

1. Describe the image you want to create in a few words.
   Example: Create an image of an Obsidian igneous rock.
2. Add in the style or type of image that you want. Some examples could be photorealistic, impressionist, watercolors, flat vector art, chalk drawing, layered paper image, icon, or anything else you can think of.
   Example: Create a photorealistic image of an obsidian rock.

Once Copilot has given you an output (typically between 1 and 4 images), you can then refine your prompt based on that output.

3. Determine what details about the picture you don’t like. This could be the background, the image style, the coloring, or something else.
4. Type in a new prompt that requests that those details are changed.
   Example prompt: Can you make the obsidian sitting on a table and give it a neutral background?

This process can be repeated as many times as necessary to create the images that you need.

While copilot can be useful, it does have some limitations. For example, it frequently returns odd computer-generated 3D art, cartoonish images, fictional language text, disproportionate images, and so on. It also has difficulty generating text in images, and often returns fictional language text.

Adobe Firefly

If you decide to use Adobe Firefly, go to Adobe Firefly and log in with your Adobe credentials. If you’ve never used Adobe Firefly before, you can review the tutorials on the homepage to get an idea of some of the available features before beginning. Once you are ready, type your prompt into the text field and click on generate. Keep in mind that each time you enter a prompt and select “generate,” it will use 1 of your 525 monthly credits.

Once Firefly begins generating, it will take you to a new screen with advanced tools on the righthand side. These tools let you choose your content type (photo or art), style, and more. Another especially useful feature, prompt suggestions, can be turned on with a toggle in the prompt entry box. Once you’ve turned this feature on, prompt suggestions will appear as you type in the prompt box. This can be handy if you need help refining your prompt. Just remember that each “generate” click results in a credit being used.

Adobe Firefly AI in Other Adobe Products

You can use Firefly with other Adobe Cloud products, such as Photoshop, Illustrator, and Stock. These products can give you more options (generative fill, generative expand, etc.) for your images, depending on how well you know them. Please see the Adobe Welcome to Generative AI page for more details.

Other Use Cases [Accessibility]

You can also use Copilot as an alt text generator for images that you already have. Alt text can occasionally be difficult to write, and asking copilot for help can give you a good starting point. As always, be sure to carefully check the AI output for accuracy before using it.

Example Prompts:

• Create 3 alternative text choices for this image. Please keep the descriptions to 1-2 sentences.
• Create 3 alternative text choices for this image, particularly describing the clouds. Please keep the descriptions to 1 sentence.
• Create 3 alternative text choices for this image describing the image in general, particularly what is highlighted in the red rectangle. Please keep the descriptions to 1-2 sentences.

After generating the alternative text, you can adjust it yourself or paste the alt text into the prompt box and use prompts to ask Copilot to refine and adjust it for you. Be sure that the final product is usable and useful alt text for the image.

Example Prompts:

• Analyze the following passage for the principles of good alternative text and provide suggestions for how I could improve it.
• Edit the following alternative text passage for concision.
• Edit the following alternative text passage so that it’s in active voice.
• Create 3 alternative text choices for this image, particularly describing the clouds. Please keep the descriptions to 1 sentence.

Conclusion

To sum up, Copilot and Adobe Firefly are two different AI tools that can help you create visual content based on text prompts. Copilot is more suitable for simple and conceptual images and alternative text, while Adobe Firefly offers more control and realism. However, both tools have their limitations and challenges, such as quality, accuracy, and prompt engineering challenges.